A quick peek at SciPy
"There are a great many people in the country today, who through no fault of their own, are sane." -- The Silly Vicar Sketch
| Author: | Stéfan van der Walt |
|---|---|
| Date: | 14/02/2007 |
- Lecture 1: The Python language
- Lecture 2: Python for scientific use
Pieces of the Puzzle
- Numpy -- N-dimensional array functionality
- Scipy -- Scientific toolkit
- Matplotlib -- Plotting
- IPython -- Console
- Other modules/applications
- Visual (3D animations)
- wxPython (Graphical user interfaces)
- mayavi (visual data inspection)
NumPy
- N-dimensional array
- Basic linear algebra functions
- Eigenvalues
- Matrix factorisation
- Linear system solver
- SVD
- and more
- Discrete fourier transform
- Random number generator
- Compatibility layers for Numeric, numarray
Constructing arrays
Provides a new class -- N-dimensional arrays.
import numpy as N
x = N.array([1,2,3])
x = N.array([1,2,3])
x = N.array([1,2,3], dtype=int)
x = N.zeros((50,50,3), dtype=float)
Note the concept of a data type -- int, float, str, object, record.
>>> import numpy as N
>>> x = N.array([1,2,3])
>>> y = N.array([3,2,1])
>>> x+y
array([4, 4, 4])
>>> x = N.array([1,2,3])
>>> y = N.array([3,2,1])
>>> x+y
array([4, 4, 4])
Array Slicing
Way to get hold of sub-arrays:
x = N.random.rand((5,5))
x[0,0]
x[0,:]
x[0,0]
x[0,:]
x = N.random.rand((5,5,3))
x[0]
x[1,...]
x[0]
x[1,...]
Array shape and type manipulation
Change shape:
x = N.arange(12)
x.reshape((4,3))
x.flat
x.reshape((4,3))
x.flat
Change data type:
x.astype(float)
Extending numpy
- Needed under special circuimstances only
- Profile first and/or identify bottlenecks
- f2py for Fortran code
- ctypes, pyrex, SWIG, boost-python for C code
- Weave for inline C code
- Native C API
SciPy
- Statistics
- Optimisation
- Numeric integration
- Linear algebra
- Fourier transforms
- Signal and image processing
- Genetic algorithms
- ODE solvers
- Special functions (Bessel etc.)
How to find your way around SciPy
- Introspection using IPython
- pydoc -g
- Read the source, Luke!
Documentation:
Pretty pictures with matplotlib
Example plot
- History of matplotlib
- High-quality graphs
- Supports LaTeX captions
- Roughly follows MATLAB (R) interface
- Can be embedded in applications
- Generates output in various formats, including png, jpeg, eps, svg
- Matplotlib tutorial
Demos
- SVD geometry
- Log polar transform
- Radial distortion
- Visual python
- FFT2
Testing your code
Peace of mind at the press of a button...
def hello():
return 'hello'
if __name__ == '__main__':
from numpy.testing import *
class test_hello(NumpyTestCase):
def test_string(self):
assert_equal(hello(), 'hello')
NumpyTest().test()
return 'hello'
if __name__ == '__main__':
from numpy.testing import *
class test_hello(NumpyTestCase):
def test_string(self):
assert_equal(hello(), 'hello')
NumpyTest().test()
Installation
Windows - Enthought's enstaller packages mirrored on archive
- New version before the end of the week for PyCon2007
- Alternatively: Enthon full package
- For optimal performance, build it yourself
Linux - Build from source -- should be as simple as
python setup.py install --prefix ${HOME}- Ubuntu: numpy r1.0.1 (Feisty), numpy r1.0rc1 (Edgy), dapper
- Ubuntu/Debian: http://deb-scipy.alioth.debian.org/apt/